Research On Multi-scale Non-rigid 3D Shape Matching Method

With the wide application of 3D shape,shape matching,as a basic research problem in the field of 3D shape analysis,has become a hot topic.The problem is to find the point correspondence between two shapes.At present,there are relatively mature algorithms for matching between rigid shapes.However,due to the deformation complexity of non-rigid shapes,there are still many problems in the existing non-rigid shape matching algorithms.In this paper,the approximate matching problem of non-rigid shapes with intrinsic distance between corresponding points during deformation is studied.In this paper,a multi-scale non-rigid shape matching method is proposed based on the consistency point drift algorithm.The proposed algorithm uses multiple simple matches in spectral space to obtain more abundant matching information,and obtains the dense matching between shapes more quickly.In the preprocessing operation of the shape subsampling,the Kcenter point clustering sampling based on LB operator is used to make the acquired sampling points more concentrated in the places with obvious shape features,so that more matching pairs are reserved in the places with obvious shape features.Then,the consistency point drift algorithm is used in spectral space to quickly obtain the matching results in a single scale space,and then the rich matching pair information is fused to obtain the initial rough matching between the shapes.In the initial rough matching,most of the correct matching pairs in the matching results of each scale are retained,but some of the wrong matches are also included.The thermal core features were used to extract the shape feature points and the approximate equidistant method was used to eliminate the error matches from the initial rough matching results in the local range of the feature points,thus the rough matching of the shape pair was obtained.Finally,the shape is gradually up-sampled by multi-scale matching method and the matching results from coarse to thin are calculated by grouping,until the dense matching results among the shapes are finally obtained.The method in this paper combines the clustering sampling based on LB operator and the farthest point sampling in the matching process,so that the coarse matching results contain more shape feature information,and at the same time,the dense matching results of shapes can be obtained more evenly and quickly.Due to the fact that a small number of sampling points are used to match in grouping matching and the consistency point drift algorithm has a fast running efficiency for a small number of point sets,this method can quickly calculate the coarse to fine matching results of shape pairs.It can be found that the proposed method is robust for shape matching of different categories by conducting multiple experiments on TOSCA database with other methods,and can improve operational efficiency to a certain extent when similar matching results are achieved.